What Internet Speed Do You Need for Day Trading? Latency, Not Mbps

For day trading you need far less internet speed than any broadband plan sells: CQG publishes about 20K bytes per second sustained per user, which is 0.16 Mbps, about 1% of the 15 Mbps Netflix recommends for one 4K stream. Bandwidth (the Mbps a plan advertises) was never the limit: latency (the round trip to a server and back), its behaviour under load, and packet loss decide whether your platform keeps up.

Key takeaway

CQG publishes about 20K bytes per second sustained per user, which works out to 0.16 Mbps, and even its optimal burst figure of 4.48 Mbps is under a third of the 15 Mbps Netflix recommends for 4K. Almost any working broadband carries a futures feed. What decides lag is packet loss, latency under load and distance: the fibre floor to CME in Aurora is about 12 ms round trip from New York and 64 ms from London, and no plan upgrade removes it.

How much bandwidth does a futures trading platform use?

Very little: CQG's published figures work out to 0.16 Mbps sustained and 0.96 Mbps with bursts, and even its optimal burst figure, 4.48 Mbps per user, is under a third of one 4K Netflix stream.

The CQG Integrated Client technical specifications, as of October 2026, say: "CQG software requires approximately 20K bytes/sec sustained and an average 100K bytes/sec of additional bandwidth per user for burst mode." They give optimal figures of 60K and 500K bytes per second, and each additional monitor needs roughly the same again.

CQG case	Bytes per second	Mbps	Share of one 4K stream (15 Mbps)
Sustained	20,000	0.16	1.1%
Sustained plus burst	120,000	0.96	6.4%
Optimal sustained	60,000	0.48	3.2%
Optimal sustained plus burst	560,000	4.48	29.9%

Computed by Phoenix Technologies from CQG's figures and the 4K figure on Netflix's recommended speeds page; method: Mbps = bytes per second x 8 / 1,000,000, and burst cases add burst to sustained. Reading K as 1,024 moves every result by 2.4%.

Set against a real plan, the feed nearly vanishes: the 467 Mbps weighted average advertised download that the FCC's Thirteenth Measuring Broadband America report found across 10 US ISPs in 2022. In the chart below, the plan bar is about 2,900 times the sustained feed bar. Four monitors at the optimal burst figure, 17.92 Mbps, still use under 4% of it.

Even CQG's optimal burst figure is under a third of one 4K stream, so a lagging platform is almost never short of Mbps.

We found no published figure from Rithmic (compared in CQG versus Rithmic), Tradovate, ProjectX or dxFeed. Sierra Chart's bandwidth usage notes decline to give one, because usage depends on the service, symbols, market activity and historical downloads, and suggest measuring your own in Windows Task Manager or Resource Monitor. Some feeds also bundle updates, as feed conflation and throttling explains.

Myth: you need 25 Mbps, or 100 Mbps, to day trade. False. The guides we checked that give these minimums cite no source; 25 Mbps is about 156 times CQG's sustained figure and 100 Mbps about 625 times. Starlink's testing found web pages stop loading faster beyond around 10 Mbps, and the FCC reports that "Beyond 250 Mbps, the webpage download times decreased only by minor amounts."

Which internet numbers matter for trading if not Mbps?

Four numbers matter more than download speed: idle latency, latency under load, jitter (how much latency varies from packet to packet) and packet loss (the share of packets that never arrive). Distance and technology set the first; your own line sets the other three.

The FCC splits round-trip latency into a fixed part, set by distance, transmission speed and technology, and "a variable component due to potential variability in routes and queuing delay that could increase as the network path congests with traffic." More Mbps cannot shorten the distance, and helps the variable part only when other traffic fills your line.

That case is bufferbloat (router queues that hold packets back when the line is full). In the FCC's 2022 measurements "the latency under downstream traffic load generally is significantly higher than idle latency", most of all on DSL. Read by eye from its Chart 7, loaded latency reached roughly 180 to 720 ms on DSL under upload load and up to about 380 ms on cable, while fibre stayed under about 100 ms. Some ISPs have since added active queue management (AQM), the FCC notes.

On packet loss, the FCC writes that "The 1% standard for packet loss is commonly accepted as the point at which highly interactive applications such as VoIP experience significant degradation in quality". It also notes that most enterprise service agreements guarantee 0.1% to 0.3%. For jitter, none of the sources we checked sets a trading threshold, so watch the spread between your best and worst ping during the session.

Myth: a faster plan lowers your latency to the exchange. Partly true. Switching technology changes idle latency, from 23 to 34 ms on DSL to 7 to 14 ms on fibre in the FCC data, and a bigger pipe can shorten queues if your household fills the smaller one; the FCC also notes that higher speed tiers generally have lower idle latency. A higher tier on the same line moves neither the distance nor the route, which set most of the delay from outside Chicago.

What is the latency from my city to CME in Aurora?

At best about 0.5 ms from downtown Chicago, 11.9 ms from New York, 64.0 ms from London and 148.2 ms from Sydney, round trip through fibre. Those are physical floors from great-circle distance; measured data-centre pings to Chicago run 1.24 to 1.57 times the floor, and a home line adds its own access latency on top.

According to Databento (December 2025), "The primary matching engine for CME Group's futures and options markets is housed in CyrusOne Aurora I", at 2905 E. Diehl Road, and it remains in Aurora as of 2026.

Light in optical fibre covers roughly 200 km per millisecond, since glass has a refractive index of about 1.5, per High Performance Browser Networking. Twice the distance divided by that speed is the shortest possible round trip.

City	Great-circle km to Aurora	Fibre floor to Aurora, round trip	Measured ping to Chicago	Measured vs floor to Chicago
Chicago (downtown)	52	0.5 ms	not measured	not measured
New York	1,195	11.9 ms	about 18 ms	1.57x
London	6,397	64.0 ms	91.5 ms	1.44x
Frankfurt	7,007	70.1 ms	about 107 ms	1.54x
Sydney	14,824	148.2 ms	about 185 ms	1.24x
Singapore	15,066	150.7 ms	about 218 ms	1.45x

Computed by Phoenix Technologies; method: haversine great-circle distance from city-centre coordinates to approximate coordinates for 2905 E. Diehl Road, floor = 2 x km / 200 km per ms. The measured column is WonderNetwork's server-to-server pings to Chicago, 15 to 29 October 2026, so its ratio uses the floor to Chicago. At 204 km per ms the floors fall by about 2%.

From Europe or Asia, the route's detour alone costs more than the idle-latency gap between fibre and cable. London's measured ping runs 28.0 ms above its floor to Chicago, more than the gap between the best fibre and the worst cable result in the FCC data (7 ms against 24 ms).

The chart below sets each city's fibre floor to Aurora beside its measured ping to Chicago. The gap between the two is the cable's detour from the great circle.

Where you live sets a latency floor that physics fixes and cables only raise; from London it is already 64.0 ms before your home line adds anything.

Worked through for London: the great-circle distance to Aurora is 6,397 km. Twice that is 12,794 km, and 12,794 divided by 200 km per ms is 64.0 ms. WonderNetwork's London to Chicago pings averaged about 91.5 ms over that fortnight, against a 63.5 ms floor to Chicago. With an illustrative 10 ms fibre access leg on top, a London home line sits at about 101.5 ms before your platform does any work.

Myth: your speed test ping is your latency to the exchange. False. A speed test, like the FCC's measurements, pings the nearest test server, not Aurora, and no public dataset measures home-to-Aurora latency, so estimate it as access latency plus the floor above. That is why a promise of under 20 ms to the exchange fails from Europe and barely holds from New York: 11.9 ms of floor plus 7 to 14 ms of fibre access is 18.9 to 25.9 ms.

Downtown Chicago to Aurora is 52 km, a 0.5 ms floor, so any multi-millisecond gap between a server in central Chicago and one in Aurora is routing, not distance. Ask a provider for a measured ping to Aurora rather than a city name; picking a VPS location near Aurora covers how.

Which is best for day trading: fibre, cable, DSL, Starlink or 5G?

Fibre, on the measured data: the FCC found median idle latency of 7 to 14 ms on fibre, 12 to 24 ms on cable and 23 to 34 ms on DSL, and fibre held up best under load. Starlink and 5G home internet are usable but sit higher, and every figure here stops at the provider's nearest server, not at Aurora.

At peak hours in that 2022 data, 1% or greater packet loss hit 0% to 4% of DSL, 1% to 4% of cable and 0% to 3% of fibre subscribers.

Is Starlink good enough for day trading?

On Starlink's own numbers, its US median latency sits close to the DSL and cable range, with a wider tail. Its March 2024 paper Improving Starlink's Latency reports US median peak-hour latency cut from 48.5 ms to 33 ms and says: "Worst-case peak hour latency (p99) has dropped by over 60%, from over 150ms to less than 65ms." In a July 2025 network update Starlink reported a 25.7 ms US median peak-hour latency, according to The Register (July 2025); we could not open Starlink's own update page to confirm it.

Myth: Starlink's 40 to 60 ms ping rules it out. Unsupported. That range, still repeated in guides updated in 2026, is out of date, and the satellite leg itself is, in Starlink's words, "usually under 10ms for the round-trip". The real caveat: a p99 under 65 ms still leaves 1 in 100 peak-hour readings above it.

5G home internet (fixed wireless access) measured higher. According to Ookla data reported by RCR Wireless News (December 2025), Q3 2025 median latency was 50 ms for T-Mobile, 54 ms for Verizon and 67 ms for AT&T, on Ookla's Speedtest method rather than the FCC's. CQG's system specifications accept fixed wireless as a high-speed connection, so 5G is workable; a wired fibre or cable line is steadier.

The chart below compares the five technologies on idle or median latency, peak-hour tail and latency under load. The measure that separates them is latency under load, not idle.

Idle latency separates the technologies by tens of milliseconds; latency under load separates them by hundreds.

Is Wi-Fi okay for day trading, or do I need Ethernet?

Use Ethernet. Two platform vendors put it in writing: CQG's system specifications say "Wireless Internet connections are not recommended, especially broadband over a Wi-Fi card.", and Sierra Chart's help page on data falling behind says "It is recommended to always use a wired connection when possible."

The reason is not speed. Sierra Chart explains that wireless can cut available bandwidth and raise transmission latency, and that lost packets are resent, which delays the data.

Myth: Wi-Fi is fine if the speed test looks good. False. A speed test reports throughput over a few seconds, while a feed suffers from single lost or late packets. No public study measures Wi-Fi loss on trading platforms, so test your own: run a long ping over Wi-Fi, then over a cable, and compare the worst readings and the loss count.

How much does internet latency matter for a manual trader?

For a manual trader on a quiet US line, the network is roughly 4% to 13% of the loop from seeing a price to the order's round trip; your own reaction is the rest. That changes on a trans-Atlantic route (about 34% to 36%) or a loaded cable line (about 66% to 68%), and a faster plan fixes neither.

Jain and colleagues (2015), in the International Journal of Applied and Basic Medical Research, cite a typical mean simple reaction time of about 180 to 200 ms to a visual stimulus. That is a lower bound: deciding whether to buy or sell is a choice, which is slower. The table sets it beside the network round trip for five illustrative setups.

Setup (illustrative)	Network round trip	Network share of the loop
Chicago, fibre	7.5 to 14.5 ms	3.6% to 7.5%
New York, fibre	18.9 to 25.9 ms	8.6% to 12.6%
US Starlink median, near Chicago	26.2 ms	11.6% to 12.7%
London, fibre	101.5 ms	33.7% to 36.1%
US cable under download load	about 380 ms	65.5% to 67.9%

Computed by Phoenix Technologies; method: network share = network / (network + reaction), using a 200 ms reaction with the fastest network and 180 ms with the slowest. Inputs: FCC idle fibre plus the fibre floor, Starlink's 25.7 ms plus 0.5 ms, an illustrative 10 ms access leg plus 91.5 ms for London, and about 380 ms read from the FCC chart. Exchange matching time is left out: no source gives it for a retail path.

A busy cable line in Chicago can cost more time than living in London.

The chart below draws one click to scale on a single time axis: a 180 to 200 ms block for the eye and hand, then the network block for each setup in the table. The Chicago fibre and Starlink blocks are slivers; the loaded cable block is about twice the human block.

On a quiet US line your own reaction is nearly the whole loop; queues on a busy line cost more than an ocean crossing.

Myth: network milliseconds decide a manual trader's fills. Partly true, as the table shows. No public study links a retail trader's home latency or packet loss to slippage or profit and loss; what does move fills is in where futures slippage comes from, and the later stages of an automated order in the end-to-end latency budget.

Why does my trading platform lag or disconnect, and what do I fix first?

Check packet loss first, then latency under load, then Wi-Fi, then your PC; distance comes next, and a faster plan comes last. Each of the first four can freeze a feed on a line with plenty of Mbps.

Myth: a little packet loss does not matter because TCP resends it. False. Sierra Chart says the connection "has to be perfect in both directions (sending and receiving) with no packet loss" to avoid a stopped or delayed feed, and that the resending is itself the problem: "This retransmission causes data delays".

Myth: lag on the chart always means bad internet. False. The same Sierra Chart page lists a busy CPU, too many charts, antivirus and firewall software among the causes of data falling behind. If a price looks wrong rather than late, see why chart and broker prices differ; for the first signs of a dead connection, how to spot a frozen feed.

The order of tests forms a short tree: from freezes or disconnects it branches on packet loss, on latency under load, on Wi-Fi versus cable, and on CPU and chart count. Only if all four pass and your floor to Aurora is too slow for your style does it end at a server near Aurora; every other leaf says a faster plan will not help.

Four cheap checks come before any purchase, and only distance justifies moving to a server; no branch ends in buying more Mbps.

When is a VPS or a backup line the fix?

A VPS (a rented server in a data centre) near Aurora is the fix only when distance is the problem, because it moves the software that sends orders next to the exchange. Whether you need a VPS at all depends on where your software runs, and prop firms set their own rules (see prop firm VPN and VPS rules). A server-based copier such as Thor runs 24/7 on cloud servers, so follower orders never leave from your home line.

A backup line protects you against an outage, not against slowness, so it should differ from the main line in technology and path: a 5G or Starlink connection behind a cable or fibre line, for example. Test it in a quiet session and learn how your platform reconnects before you need it; if copies stop after a switch, use the copier diagnostic by symptom.

Your decision rule:

If you see any sustained packet loss during the session, fix it before anything else: the cable, the router, then the ISP.
If ping jumps from tens to hundreds of ms while someone at home uploads or streams, you have bufferbloat: ask your ISP or check your router for active queue management, or keep big transfers off the line while you trade.
If you trade over Wi-Fi, run a cable.
If lag grows with open charts or CPU load, fix the PC, not the line.
If your line holds about 4.48 Mbps per monitor (CQG's optimal burst figure) without loss, a faster plan buys you nothing for trading.
If your floor to Aurora is too slow for an automated or copy setup, move the software to a server near Aurora instead of upgrading your plan.

Go deeper

Frequently asked questions

Does upload speed matter for day trading?

Upload congestion matters more than upload speed. No vendor we checked publishes a separate upload figure, and CQG's whole per-user figure is 0.16 Mbps sustained. In the FCC's 2022 data, DSL latency under upload load reached roughly 180 to 720 ms, read from its chart, so keep large uploads such as cloud backups off the line while you trade.

Is 100 Mbps enough for day trading?

Yes, many times over. 100 Mbps is about 625 times CQG's sustained per-user figure and covers four monitors at its optimal burst figure, 17.92 Mbps, several times. If a 100 Mbps line still lags, look at packet loss, queuing, Wi-Fi or the PC.

Do more monitors or charts use more bandwidth?

Yes. CQG says each additional monitor needs roughly the same bandwidth again, so four monitors at its optimal burst figure come to about 17.92 Mbps. Sierra Chart adds that usage depends on symbol count, market activity and historical downloads, so loading history and adding symbols both raise it. Measure your own rate in Task Manager or Resource Monitor.

Can a VPN lower my latency to CME?

It cannot beat the distance floor, because no fibre route is shorter than the great circle; from London that floor is 64.0 ms. A VPN adds a server to your route, which can help or hurt depending on your ISP's routing, and only a test shows which. Prop firms set their own VPN rules, so check the firm's current rules first.

Does it matter where my broker's or data vendor's servers are?

Yes, because your order passes through the vendor's own servers on its way to CME. Retail vendors do not publish where their order gateways and risk servers sit in a comparable way, so the true path of a retail order is unknown from outside. A ping to your vendor's server, taken from your platform, is the closest measurement you can make.

Will CME's Google Cloud plan change latency from home?

Not the destination, as of 2026. Databento reports that CME Group and Google Cloud announced in 2024 a plan for a new Google Cloud private region and colocation facility in Aurora, and that the Globex matching engine remains in Aurora. With both in Aurora, the distance floors from your city still apply.

Why does my feed fall behind only at the open or on news?

Bursts raise the data rate: CQG budgets an extra 100K bytes per second per user for burst mode, taking its figure from 0.16 Mbps to 0.96 Mbps. That still fits any broadband, so a feed that falls behind at the open more likely points to the PC, too many charts or packet loss, causes Sierra Chart lists.

Is a business internet plan better for day trading?

It can buy a written packet loss guarantee rather than lower latency: the FCC notes that most enterprise service agreements guarantee 0.1% to 0.3% packet loss. If a standard plan already shows no loss during your sessions, the contract changes little. Distance to Aurora is the same on either plan.